ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.89 by root, Thu Jul 14 22:36:17 2011 UTC vs.
Revision 1.149 by root, Tue Feb 20 06:54:47 2018 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
50#include "xthread.h"
51
52#include <errno.h> 47#include <errno.h>
53#include <stddef.h> 48#include <stddef.h>
54#include <stdlib.h> 49#include <stdlib.h>
55#include <string.h> 50#include <string.h>
56#include <errno.h> 51#include <errno.h>
67#endif 62#endif
68 63
69#ifndef ECANCELED 64#ifndef ECANCELED
70# define ECANCELED EDOM 65# define ECANCELED EDOM
71#endif 66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
72 74
73static void eio_destroy (eio_req *req); 75static void eio_destroy (eio_req *req);
74 76
75#ifndef EIO_FINISH 77#ifndef EIO_FINISH
76# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
95 97
96#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) 98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
97 99
98#ifdef _WIN32 100#ifdef _WIN32
99 101
102 #undef PAGESIZE
100 #define PAGESIZE 4096 /* GetSystemInfo? */ 103 #define PAGESIZE 4096 /* GetSystemInfo? */
101 104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
102 #ifdef EIO_STRUCT_STATI64 107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
103 #define stat(path,buf) _stati64 (path,buf) 109 #define stat(path,buf) _stati64 (path,buf)
104 #define fstat(fd,buf) _fstati64 (path,buf) 110 #define fstat(fd,buf) _fstati64 (fd,buf)
105 #endif 111 #endif
106 #define lstat(path,buf) stat (path,buf) 112 #define lstat(path,buf) stat (path,buf)
107 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
108 #define mkdir(path,mode) _mkdir (path) 114 #define mkdir(path,mode) _mkdir (path)
109 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
110 116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
111 #define chown(path,uid,gid) EIO_ENOSYS () 125 #define chown(path,uid,gid) EIO_ENOSYS ()
112 #define fchown(fd,uid,gid) EIO_ENOSYS () 126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
113 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
114 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ 128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
115 #define mknod(path,mode,dev) EIO_ENOSYS () 129 #define mknod(path,mode,dev) EIO_ENOSYS ()
116 #define sync() EIO_ENOSYS () 130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
117 200
118 /* we could even stat and see if it exists */ 201 /* we could even stat and see if it exists */
119 static int 202 static int
120 symlink (const char *old, const char *neu) 203 symlink (const char *old, const char *neu)
121 { 204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
122 if (CreateSymbolicLink (neu, old, 1)) 213 if (CreateSymbolicLink (neu, old, flags))
123 return 0; 214 return 0;
124 215 #endif
125 if (CreateSymbolicLink (neu, old, 0))
126 return 0;
127 216
128 return EIO_ERRNO (ENOENT, -1); 217 return EIO_ERRNO (ENOENT, -1);
129 } 218 }
130 219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
233
131#else 234#else
132 235
236 #include <sys/ioctl.h>
133 #include <sys/time.h> 237 #include <sys/time.h>
134 #include <sys/select.h> 238 #include <sys/select.h>
135 #include <sys/statvfs.h>
136 #include <unistd.h> 239 #include <unistd.h>
137 #include <utime.h>
138 #include <signal.h> 240 #include <signal.h>
139 #include <dirent.h> 241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
140 251
141 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
142 #include <sys/mman.h> 253 #include <sys/mman.h>
143 #endif 254 #endif
144 255
256 #define D_NAME(entp) entp->d_name
257
145 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
146 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
147 #define _DIRENT_HAVE_D_TYPE /* sigh */ 260 #define _DIRENT_HAVE_D_TYPE /* sigh */
148 #define D_INO(de) (de)->d_fileno 261 #define D_INO(de) (de)->d_fileno
149 #define D_NAMLEN(de) (de)->d_namlen 262 #define D_NAMLEN(de) (de)->d_namlen
150 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
151 #define D_INO(de) (de)->d_ino 264 #define D_INO(de) (de)->d_ino
162 275
163 #ifndef EIO_STRUCT_DIRENT 276 #ifndef EIO_STRUCT_DIRENT
164 #define EIO_STRUCT_DIRENT struct dirent 277 #define EIO_STRUCT_DIRENT struct dirent
165 #endif 278 #endif
166 279
280#endif
281
282#if HAVE_UTIMES
283# include <utime.h>
284#endif
285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
167#endif 288#endif
168 289
169#if HAVE_SENDFILE 290#if HAVE_SENDFILE
170# if __linux 291# if __linux
171# include <sys/sendfile.h> 292# include <sys/sendfile.h>
179# else 300# else
180# error sendfile support requested but not available 301# error sendfile support requested but not available
181# endif 302# endif
182#endif 303#endif
183 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
184#ifndef D_TYPE 310#ifndef D_TYPE
185# define D_TYPE(de) 0 311# define D_TYPE(de) 0
186#endif 312#endif
187#ifndef D_INO 313#ifndef D_INO
188# define D_INO(de) 0 314# define D_INO(de) 0
189#endif 315#endif
190#ifndef D_NAMLEN 316#ifndef D_NAMLEN
191# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
192#endif 318#endif
193 319
194/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
195#ifndef NAME_MAX 321#ifndef NAME_MAX
196# define NAME_MAX 4096 322# define NAME_MAX 4096
197#endif 323#endif
198 324
199/* used for readlink etc. */ 325/* used for readlink etc. */
200#ifndef PATH_MAX 326#ifndef PATH_MAX
201# define PATH_MAX 4096 327# define PATH_MAX 0
202#endif 328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef EIO_PATH_MIN
335# define EIO_PATH_MIN 8160
336#endif
337
338#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
203 339
204/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
205#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
206 342
207#define dBUF \ 343#define dBUF \
208 char *eio_buf; \
209 ETP_WORKER_LOCK (self); \
210 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
211 ETP_WORKER_UNLOCK (self); \
212 errno = ENOMEM; \ 345 errno = ENOMEM; \
213 if (!eio_buf) \ 346 if (!eio_buf) \
214 return -1; 347 return -1
215 348
216#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
351
352/*****************************************************************************/
353
354struct etp_tmpbuf;
355
356#if _POSIX_VERSION >= 200809L
357 #define HAVE_AT 1
358 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
359 #ifndef O_SEARCH
360 #define O_SEARCH O_RDONLY
361 #endif
362#else
363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
365#endif
366
367struct eio_pwd
368{
369#if HAVE_AT
370 int fd;
371#endif
372 int len;
373 char str[1]; /* actually, a 0-terminated canonical path */
374};
375
376/*****************************************************************************/
217 377
218#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
219#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
220 380
381#define ETP_TYPE_QUIT -1
382#define ETP_TYPE_GROUP EIO_GROUP
383
384static void eio_nop_callback (void) { }
385static void (*eio_want_poll_cb)(void) = eio_nop_callback;
386static void (*eio_done_poll_cb)(void) = eio_nop_callback;
387
388#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
389#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
390
221struct etp_worker; 391struct etp_worker;
222
223#define ETP_REQ eio_req 392#define ETP_REQ eio_req
224#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
225static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
226#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
227static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
228#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
229 398
230#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
231 if (wrk->dbuf) \
232 { \
233 free (wrk->dbuf); \
234 wrk->dbuf = 0; \
235 } \
236 \
237 if (wrk->dirp) \
238 { \
239 closedir (wrk->dirp); \
240 wrk->dirp = 0; \
241 }
242 400
243#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
244 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
245 DIR *dirp;
246
247/*****************************************************************************/
248
249#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
250
251/* calculate time difference in ~1/EIO_TICKS of a second */
252ecb_inline int
253tvdiff (struct timeval *tv1, struct timeval *tv2)
254{
255 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
256 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
257}
258
259static unsigned int started, idle, wanted = 4;
260
261static void (*want_poll_cb) (void);
262static void (*done_poll_cb) (void);
263
264static unsigned int max_poll_time; /* reslock */
265static unsigned int max_poll_reqs; /* reslock */
266
267static volatile unsigned int nreqs; /* reqlock */
268static volatile unsigned int nready; /* reqlock */
269static volatile unsigned int npending; /* reqlock */
270static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
271static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
272
273static xmutex_t wrklock;
274static xmutex_t reslock;
275static xmutex_t reqlock;
276static xcond_t reqwait;
277
278#if !HAVE_PREADWRITE
279/*
280 * make our pread/pwrite emulation safe against themselves, but not against
281 * normal read/write by using a mutex. slows down execution a lot,
282 * but that's your problem, not mine.
283 */
284static xmutex_t preadwritelock = X_MUTEX_INIT;
285#endif
286
287typedef struct etp_worker
288{
289 /* locked by wrklock */
290 struct etp_worker *prev, *next;
291
292 xthread_t tid;
293
294 /* locked by reslock, reqlock or wrklock */
295 ETP_REQ *req; /* currently processed request */
296
297 ETP_WORKER_COMMON
298} etp_worker;
299
300static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
301
302#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
303#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
304
305/* worker threads management */
306
307static void ecb_cold
308etp_worker_clear (etp_worker *wrk)
309{
310 ETP_WORKER_CLEAR (wrk);
311}
312
313static void ecb_cold
314etp_worker_free (etp_worker *wrk)
315{
316 wrk->next->prev = wrk->prev;
317 wrk->prev->next = wrk->next;
318
319 free (wrk);
320}
321
322static unsigned int
323etp_nreqs (void)
324{
325 int retval;
326 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
327 retval = nreqs;
328 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
329 return retval;
330}
331
332static unsigned int
333etp_nready (void)
334{
335 unsigned int retval;
336
337 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
338 retval = nready;
339 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
340
341 return retval;
342}
343
344static unsigned int
345etp_npending (void)
346{
347 unsigned int retval;
348
349 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
350 retval = npending;
351 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
352
353 return retval;
354}
355
356static unsigned int
357etp_nthreads (void)
358{
359 unsigned int retval;
360
361 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
362 retval = started;
363 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
364
365 return retval;
366}
367
368/*
369 * a somewhat faster data structure might be nice, but
370 * with 8 priorities this actually needs <20 insns
371 * per shift, the most expensive operation.
372 */
373typedef struct {
374 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
375 int size;
376} etp_reqq;
377
378static etp_reqq req_queue;
379static etp_reqq res_queue;
380
381static int ecb_noinline
382reqq_push (etp_reqq *q, ETP_REQ *req)
383{
384 int pri = req->pri;
385 req->next = 0;
386
387 if (q->qe[pri])
388 {
389 q->qe[pri]->next = req;
390 q->qe[pri] = req;
391 }
392 else
393 q->qe[pri] = q->qs[pri] = req;
394
395 return q->size++;
396}
397
398static ETP_REQ * ecb_noinline
399reqq_shift (etp_reqq *q)
400{
401 int pri;
402
403 if (!q->size)
404 return 0;
405
406 --q->size;
407
408 for (pri = ETP_NUM_PRI; pri--; )
409 {
410 eio_req *req = q->qs[pri];
411
412 if (req)
413 {
414 if (!(q->qs[pri] = (eio_req *)req->next))
415 q->qe[pri] = 0;
416
417 return req;
418 }
419 }
420
421 abort ();
422}
423
424static void ecb_cold
425etp_thread_init (void)
426{
427#if !HAVE_PREADWRITE
428 X_MUTEX_CREATE (preadwritelock);
429#endif
430 X_MUTEX_CREATE (wrklock);
431 X_MUTEX_CREATE (reslock);
432 X_MUTEX_CREATE (reqlock);
433 X_COND_CREATE (reqwait);
434}
435
436static void ecb_cold
437etp_atfork_prepare (void)
438{
439}
440
441static void ecb_cold
442etp_atfork_parent (void)
443{
444}
445
446static void ecb_cold
447etp_atfork_child (void)
448{
449 ETP_REQ *prv;
450
451 while ((prv = reqq_shift (&req_queue)))
452 ETP_DESTROY (prv);
453
454 while ((prv = reqq_shift (&res_queue)))
455 ETP_DESTROY (prv);
456
457 while (wrk_first.next != &wrk_first)
458 {
459 etp_worker *wrk = wrk_first.next;
460
461 if (wrk->req)
462 ETP_DESTROY (wrk->req);
463
464 etp_worker_clear (wrk);
465 etp_worker_free (wrk);
466 }
467
468 started = 0;
469 idle = 0;
470 nreqs = 0;
471 nready = 0;
472 npending = 0;
473
474 etp_thread_init ();
475}
476
477static void ecb_cold
478etp_once_init (void)
479{
480 etp_thread_init ();
481 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
482}
483
484static int ecb_cold
485etp_init (void (*want_poll)(void), void (*done_poll)(void))
486{
487 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
488
489 pthread_once (&doinit, etp_once_init);
490
491 want_poll_cb = want_poll;
492 done_poll_cb = done_poll;
493
494 return 0;
495}
496
497X_THREAD_PROC (etp_proc);
498
499static void ecb_cold
500etp_start_thread (void)
501{
502 etp_worker *wrk = calloc (1, sizeof (etp_worker));
503
504 /*TODO*/
505 assert (("unable to allocate worker thread data", wrk));
506
507 X_LOCK (wrklock);
508
509 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
510 {
511 wrk->prev = &wrk_first;
512 wrk->next = wrk_first.next;
513 wrk_first.next->prev = wrk;
514 wrk_first.next = wrk;
515 ++started;
516 }
517 else
518 free (wrk);
519
520 X_UNLOCK (wrklock);
521}
522
523static void
524etp_maybe_start_thread (void)
525{
526 if (ecb_expect_true (etp_nthreads () >= wanted))
527 return;
528
529 /* todo: maybe use idle here, but might be less exact */
530 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
531 return;
532
533 etp_start_thread ();
534}
535
536static void ecb_cold
537etp_end_thread (void)
538{
539 eio_req *req = calloc (1, sizeof (eio_req));
540
541 req->type = -1;
542 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
543
544 X_LOCK (reqlock);
545 reqq_push (&req_queue, req);
546 X_COND_SIGNAL (reqwait);
547 X_UNLOCK (reqlock);
548
549 X_LOCK (wrklock);
550 --started;
551 X_UNLOCK (wrklock);
552}
553
554static int
555etp_poll (void)
556{
557 unsigned int maxreqs;
558 unsigned int maxtime;
559 struct timeval tv_start, tv_now;
560
561 X_LOCK (reslock);
562 maxreqs = max_poll_reqs;
563 maxtime = max_poll_time;
564 X_UNLOCK (reslock);
565
566 if (maxtime)
567 gettimeofday (&tv_start, 0);
568
569 for (;;)
570 {
571 ETP_REQ *req;
572
573 etp_maybe_start_thread ();
574
575 X_LOCK (reslock);
576 req = reqq_shift (&res_queue);
577
578 if (req)
579 {
580 --npending;
581
582 if (!res_queue.size && done_poll_cb)
583 done_poll_cb ();
584 }
585
586 X_UNLOCK (reslock);
587
588 if (!req)
589 return 0;
590
591 X_LOCK (reqlock);
592 --nreqs;
593 X_UNLOCK (reqlock);
594
595 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
596 {
597 req->int1 = 1; /* mark request as delayed */
598 continue;
599 }
600 else
601 {
602 int res = ETP_FINISH (req);
603 if (ecb_expect_false (res))
604 return res;
605 }
606
607 if (ecb_expect_false (maxreqs && !--maxreqs))
608 break;
609
610 if (maxtime)
611 {
612 gettimeofday (&tv_now, 0);
613
614 if (tvdiff (&tv_start, &tv_now) >= maxtime)
615 break;
616 }
617 }
618
619 errno = EAGAIN;
620 return -1;
621}
622
623static void
624etp_cancel (ETP_REQ *req)
625{
626 req->cancelled = 1;
627
628 eio_grp_cancel (req);
629}
630
631static void
632etp_submit (ETP_REQ *req)
633{
634 req->pri -= ETP_PRI_MIN;
635
636 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
637 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
638
639 if (ecb_expect_false (req->type == EIO_GROUP))
640 {
641 /* I hope this is worth it :/ */
642 X_LOCK (reqlock);
643 ++nreqs;
644 X_UNLOCK (reqlock);
645
646 X_LOCK (reslock);
647
648 ++npending;
649
650 if (!reqq_push (&res_queue, req) && want_poll_cb)
651 want_poll_cb ();
652
653 X_UNLOCK (reslock);
654 }
655 else
656 {
657 X_LOCK (reqlock);
658 ++nreqs;
659 ++nready;
660 reqq_push (&req_queue, req);
661 X_COND_SIGNAL (reqwait);
662 X_UNLOCK (reqlock);
663
664 etp_maybe_start_thread ();
665 }
666}
667
668static void ecb_cold
669etp_set_max_poll_time (double nseconds)
670{
671 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
672 max_poll_time = nseconds * EIO_TICKS;
673 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
674}
675
676static void ecb_cold
677etp_set_max_poll_reqs (unsigned int maxreqs)
678{
679 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
680 max_poll_reqs = maxreqs;
681 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
682}
683
684static void ecb_cold
685etp_set_max_idle (unsigned int nthreads)
686{
687 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
688 max_idle = nthreads;
689 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
690}
691
692static void ecb_cold
693etp_set_idle_timeout (unsigned int seconds)
694{
695 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
696 idle_timeout = seconds;
697 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
698}
699
700static void ecb_cold
701etp_set_min_parallel (unsigned int nthreads)
702{
703 if (wanted < nthreads)
704 wanted = nthreads;
705}
706
707static void ecb_cold
708etp_set_max_parallel (unsigned int nthreads)
709{
710 if (wanted > nthreads)
711 wanted = nthreads;
712
713 while (started > wanted)
714 etp_end_thread ();
715}
716 403
717/*****************************************************************************/ 404/*****************************************************************************/
718 405
719static void 406static void
720grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
721{ 408{
722 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
723 { 410 {
724 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
725 412
726 EIO_FEED (grp); 413 EIO_FEED (grp);
727 414
728 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
729 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
730 { 417 {
731 grp->feed = 0; 418 grp->feed = 0;
732 break; 419 break;
733 } 420 }
734 } 421 }
741 428
742 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
743 grp_try_feed (grp); 430 grp_try_feed (grp);
744 431
745 /* finish, if done */ 432 /* finish, if done */
746 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
747 return eio_finish (grp); 434 return eio_finish (grp);
748 else 435 else
749 return 0; 436 return 0;
750} 437}
751 438
787} 474}
788 475
789void 476void
790eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
791{ 478{
792 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
793 eio_cancel (grp);
794} 480}
795 481
796void 482void
797eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
798{ 484{
799 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
800} 486}
801 487
802void 488void
803eio_submit (eio_req *req) 489eio_submit (eio_req *req)
804{ 490{
805 etp_submit (req); 491 etp_submit (EIO_POOL, req);
806} 492}
807 493
808unsigned int 494unsigned int
809eio_nreqs (void) 495eio_nreqs (void)
810{ 496{
811 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
812} 498}
813 499
814unsigned int 500unsigned int
815eio_nready (void) 501eio_nready (void)
816{ 502{
817 return etp_nready (); 503 return etp_nready (EIO_POOL);
818} 504}
819 505
820unsigned int 506unsigned int
821eio_npending (void) 507eio_npending (void)
822{ 508{
823 return etp_npending (); 509 return etp_npending (EIO_POOL);
824} 510}
825 511
826unsigned int ecb_cold 512unsigned int ecb_cold
827eio_nthreads (void) 513eio_nthreads (void)
828{ 514{
829 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
830} 516}
831 517
832void ecb_cold 518void ecb_cold
833eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
834{ 520{
835 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
836} 522}
837 523
838void ecb_cold 524void ecb_cold
839eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
840{ 526{
841 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
842} 528}
843 529
844void ecb_cold 530void ecb_cold
845eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
846{ 532{
847 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
848} 534}
849 535
850void ecb_cold 536void ecb_cold
851eio_set_idle_timeout (unsigned int seconds) 537eio_set_idle_timeout (unsigned int seconds)
852{ 538{
853 etp_set_idle_timeout (seconds); 539 etp_set_idle_timeout (EIO_POOL, seconds);
854} 540}
855 541
856void ecb_cold 542void ecb_cold
857eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
858{ 544{
859 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
860} 546}
861 547
862void ecb_cold 548void ecb_cold
863eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
864{ 550{
865 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
866} 552}
867 553
868int eio_poll (void) 554int eio_poll (void)
869{ 555{
870 return etp_poll (); 556 return etp_poll (EIO_POOL);
871} 557}
872 558
873/*****************************************************************************/ 559/*****************************************************************************/
874/* work around various missing functions */ 560/* work around various missing functions */
875 561
876#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
877# undef pread 563# define eio__close(fd) posix_close (fd, 0)
878# undef pwrite 564#else
879# define pread eio__pread 565# define eio__close(fd) close (fd)
880# define pwrite eio__pwrite
881
882static eio_ssize_t
883eio__pread (int fd, void *buf, size_t count, off_t offset)
884{
885 eio_ssize_t res;
886 off_t ooffset;
887
888 X_LOCK (preadwritelock);
889 ooffset = lseek (fd, 0, SEEK_CUR);
890 lseek (fd, offset, SEEK_SET);
891 res = read (fd, buf, count);
892 lseek (fd, ooffset, SEEK_SET);
893 X_UNLOCK (preadwritelock);
894
895 return res;
896}
897
898static eio_ssize_t
899eio__pwrite (int fd, void *buf, size_t count, off_t offset)
900{
901 eio_ssize_t res;
902 off_t ooffset;
903
904 X_LOCK (preadwritelock);
905 ooffset = lseek (fd, 0, SEEK_CUR);
906 lseek (fd, offset, SEEK_SET);
907 res = write (fd, buf, count);
908 lseek (fd, ooffset, SEEK_SET);
909 X_UNLOCK (preadwritelock);
910
911 return res;
912}
913#endif 566#endif
567
568/* close() without disturbing errno */
569static void
570silent_close (int fd)
571{
572 int saved_errno = errno;
573 eio__close (fd);
574 errno = saved_errno;
575}
914 576
915#ifndef HAVE_UTIMES 577#ifndef HAVE_UTIMES
916 578
917# undef utimes 579# undef utimes
918# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
951 613
952#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
953# undef fdatasync 615# undef fdatasync
954# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
955#endif 617#endif
618
619static int
620eio__syncfs (int fd)
621{
622 int res;
623
624#if HAVE_SYS_SYNCFS
625 res = (int)syscall (__NR_syncfs, (int)(fd));
626#else
627 res = EIO_ENOSYS ();
628#endif
629
630 if (res < 0 && errno == ENOSYS && fd >= 0)
631 sync ();
632
633 return res;
634}
956 635
957/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
958static int 637static int
959eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 638eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
960{ 639{
983} 662}
984 663
985static int 664static int
986eio__fallocate (int fd, int mode, off_t offset, size_t len) 665eio__fallocate (int fd, int mode, off_t offset, size_t len)
987{ 666{
988#if HAVE_FALLOCATE 667#if HAVE_LINUX_FALLOCATE
989 return fallocate (fd, mode, offset, len); 668 return fallocate (fd, mode, offset, len);
990#else 669#else
991 errno = ENOSYS; 670 return EIO_ENOSYS ();
992 return -1;
993#endif 671#endif
994} 672}
995 673
996#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
997# undef readahead 675# undef readahead
1010 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
1011 offset += len; 689 offset += len;
1012 todo -= len; 690 todo -= len;
1013 } 691 }
1014 692
1015 errno = 0; 693 FUBd;
694
695 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
696 /* but not for e.g. EIO or eof, so we also never fail */
1016 return count; 697 return 0;
1017} 698}
1018 699
1019#endif 700#endif
1020 701
1021/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
1022static eio_ssize_t 703static eio_ssize_t
1023eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 704eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1024{ 705{
1025 eio_ssize_t written = 0; 706 eio_ssize_t written = 0;
1026 eio_ssize_t res; 707 eio_ssize_t res;
1027 708
1028 if (!count) 709 if (!count)
1056 737
1057 /* according to source inspection, this is correct, and useful behaviour */ 738 /* according to source inspection, this is correct, and useful behaviour */
1058 if (sbytes) 739 if (sbytes)
1059 res = sbytes; 740 res = sbytes;
1060 741
1061# elif defined (__APPLE__) 742# elif defined __APPLE__
1062 off_t sbytes = count; 743 off_t sbytes = count;
1063 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1064 745
1065 /* according to the manpage, sbytes is always valid */ 746 /* according to the manpage, sbytes is always valid */
1066 if (sbytes) 747 if (sbytes)
1093 HANDLE h = TO_SOCKET (ifd); 774 HANDLE h = TO_SOCKET (ifd);
1094 SetFilePointer (h, offset, 0, FILE_BEGIN); 775 SetFilePointer (h, offset, 0, FILE_BEGIN);
1095 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1096 777
1097#else 778#else
1098 res = -1; 779 res = EIO_ENOSYS ();
1099 errno = ENOSYS;
1100#endif 780#endif
1101 781
1102 /* we assume sendfile can copy at least 128mb in one go */ 782 /* we assume sendfile can copy at least 128mb in one go */
1103 if (res <= 128 * 1024 * 1024) 783 if (res <= 128 * 1024 * 1024)
1104 { 784 {
1126 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1127 /* BSDs */ 807 /* BSDs */
1128#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 808#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1129 || errno == ENOTSUP 809 || errno == ENOTSUP
1130#endif 810#endif
811#ifdef EOPNOTSUPP /* windows */
1131 || errno == EOPNOTSUPP /* BSDs */ 812 || errno == EOPNOTSUPP /* BSDs */
813#endif
1132#if __solaris 814#if __solaris
1133 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1134#endif 816#endif
1135 ) 817 )
1136 ) 818 )
1162 844
1163 offset += cnt; 845 offset += cnt;
1164 res += cnt; 846 res += cnt;
1165 count -= cnt; 847 count -= cnt;
1166 } 848 }
849
850 FUBd;
1167 } 851 }
1168 852
1169 return res; 853 return res;
1170} 854}
1171 855
1287 971
1288/*****************************************************************************/ 972/*****************************************************************************/
1289/* requests implemented outside eio_execute, because they are so large */ 973/* requests implemented outside eio_execute, because they are so large */
1290 974
1291static void 975static void
1292eio__realpath (eio_req *req, etp_worker *self) 976eio__lseek (eio_req *req)
1293{ 977{
1294 char *rel = req->ptr1; 978 /* this usually gets optimised away completely, or your compiler sucks, */
979 /* or the whence constants really are not 0, 1, 2 */
980 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
981 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
982 : req->int2 == EIO_SEEK_END ? SEEK_END
983 : req->int2;
984
985 req->offs = lseek (req->int1, req->offs, whence);
986 req->result = req->offs == (off_t)-1 ? -1 : 0;
987}
988
989/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
990static int
991eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
992{
1295 char *res; 993 char *res;
994 const char *rel = path;
1296 char *tmp1, *tmp2; 995 char *tmp1, *tmp2;
1297#if SYMLOOP_MAX > 32 996#if SYMLOOP_MAX > 32
1298 int symlinks = SYMLOOP_MAX; 997 int symlinks = SYMLOOP_MAX;
1299#else 998#else
1300 int symlinks = 32; 999 int symlinks = 32;
1301#endif 1000#endif
1302 1001
1303 req->result = -1;
1304
1305 errno = EINVAL; 1002 errno = EINVAL;
1306 if (!rel) 1003 if (!rel)
1307 return; 1004 return -1;
1308 1005
1309 errno = ENOENT; 1006 errno = ENOENT;
1310 if (!*rel) 1007 if (!*rel)
1311 return; 1008 return -1;
1312 1009
1313 if (!req->ptr2) 1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1314 { 1011#ifdef _WIN32
1315 X_LOCK (wrklock); 1012 if (_access (rel, 4) != 0)
1316 req->flags |= EIO_FLAG_PTR2_FREE;
1317 X_UNLOCK (wrklock);
1318 req->ptr2 = malloc (PATH_MAX * 3);
1319
1320 errno = ENOMEM;
1321 if (!req->ptr2)
1322 return; 1013 return -1;
1323 }
1324 1014
1325 res = req->ptr2; 1015 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1016
1017 errno = ENAMETOOLONG;
1018 if (symlinks >= EIO_PATH_MAX * 3)
1019 return -1;
1020
1021 errno = EIO;
1022 if (symlinks <= 0)
1023 return -1;
1024
1025 return symlinks;
1026
1027#else
1326 tmp1 = res + PATH_MAX; 1028 tmp1 = res + EIO_PATH_MAX;
1327 tmp2 = tmp1 + PATH_MAX; 1029 tmp2 = tmp1 + EIO_PATH_MAX;
1328 1030
1329#if 0 /* disabled, the musl way to do things is just too racy */ 1031#if 0 /* disabled, the musl way to do things is just too racy */
1330#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1331 /* on linux we may be able to ask the kernel */ 1033 /* on linux we may be able to ask the kernel */
1332 { 1034 {
1333 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); 1035 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1334 1036
1335 if (fd >= 0) 1037 if (fd >= 0)
1336 { 1038 {
1337 sprintf (tmp1, "/proc/self/fd/%d", fd); 1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1338 req->result = readlink (tmp1, res, PATH_MAX); 1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1339 close (fd);
1340
1341 /* here we should probably stat the open file and the disk file, to make sure they still match */ 1041 /* here we should probably stat the open file and the disk file, to make sure they still match */
1042 eio__close (fd);
1342 1043
1343 if (req->result > 0) 1044 if (req->result > 0)
1344 goto done; 1045 goto done;
1345 } 1046 }
1346 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1047 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1347 return; 1048 return -1;
1348 } 1049 }
1349#endif 1050#endif
1350#endif 1051#endif
1351 1052
1352 if (*rel != '/') 1053 if (*rel != '/')
1353 { 1054 {
1354 if (!getcwd (res, PATH_MAX)) 1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1355 return; 1059 return -1;
1060
1061 if (wd == EIO_CWD)
1062 {
1063 if (!getcwd (res, EIO_PATH_MAX))
1064 return -1;
1065
1066 len = strlen (res);
1067 }
1068 else
1069 memcpy (res, wd->str, len = wd->len);
1356 1070
1357 if (res [1]) /* only use if not / */ 1071 if (res [1]) /* only use if not / */
1358 res += strlen (res); 1072 res += len;
1359 } 1073 }
1360 1074
1361 while (*rel) 1075 while (*rel)
1362 { 1076 {
1363 eio_ssize_t len, linklen; 1077 eio_ssize_t len, linklen;
1364 char *beg = rel; 1078 const char *beg = rel;
1365 1079
1366 while (*rel && *rel != '/') 1080 while (*rel && *rel != '/')
1367 ++rel; 1081 ++rel;
1368 1082
1369 len = rel - beg; 1083 len = rel - beg;
1381 1095
1382 if (beg [1] == '.' && len == 2) 1096 if (beg [1] == '.' && len == 2)
1383 { 1097 {
1384 /* .. - back up one component, if possible */ 1098 /* .. - back up one component, if possible */
1385 1099
1386 while (res != req->ptr2) 1100 while (res != tmpbuf->ptr)
1387 if (*--res == '/') 1101 if (*--res == '/')
1388 break; 1102 break;
1389 1103
1390 continue; 1104 continue;
1391 } 1105 }
1392 } 1106 }
1393 1107
1394 errno = ENAMETOOLONG; 1108 errno = ENAMETOOLONG;
1395 if (res + 1 + len + 1 >= tmp1) 1109 if (res + 1 + len + 1 >= tmp1)
1396 return; 1110 return -1;
1397 1111
1398 /* copy one component */ 1112 /* copy one component */
1399 *res = '/'; 1113 *res = '/';
1400 memcpy (res + 1, beg, len); 1114 memcpy (res + 1, beg, len);
1401 1115
1402 /* zero-terminate, for readlink */ 1116 /* zero-terminate, for readlink */
1403 res [len + 1] = 0; 1117 res [len + 1] = 0;
1404 1118
1405 /* now check if it's a symlink */ 1119 /* now check if it's a symlink */
1406 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1407 1121
1408 if (linklen < 0) 1122 if (linklen < 0)
1409 { 1123 {
1410 if (errno != EINVAL) 1124 if (errno != EINVAL)
1411 return; 1125 return -1;
1412 1126
1413 /* it's a normal directory. hopefully */ 1127 /* it's a normal directory. hopefully */
1414 res += len + 1; 1128 res += len + 1;
1415 } 1129 }
1416 else 1130 else
1417 { 1131 {
1418 /* yay, it was a symlink - build new path in tmp2 */ 1132 /* yay, it was a symlink - build new path in tmp2 */
1419 int rellen = strlen (rel); 1133 int rellen = strlen (rel);
1420 1134
1421 errno = ENAMETOOLONG; 1135 errno = ENAMETOOLONG;
1422 if (linklen + 1 + rellen >= PATH_MAX) 1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1423 return; 1137 return -1;
1424 1138
1425 errno = ELOOP; 1139 errno = ELOOP;
1426 if (!--symlinks) 1140 if (!--symlinks)
1427 return; 1141 return -1;
1428 1142
1429 if (*tmp1 == '/') 1143 if (*tmp1 == '/')
1430 res = req->ptr2; /* symlink resolves to an absolute path */ 1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1431 1145
1432 /* we need to be careful, as rel might point into tmp2 already */ 1146 /* we need to be careful, as rel might point into tmp2 already */
1433 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1434 tmp2 [linklen] = '/'; 1148 tmp2 [linklen] = '/';
1435 memcpy (tmp2, tmp1, linklen); 1149 memcpy (tmp2, tmp1, linklen);
1437 rel = tmp2; 1151 rel = tmp2;
1438 } 1152 }
1439 } 1153 }
1440 1154
1441 /* special case for the lone root path */ 1155 /* special case for the lone root path */
1442 if (res == req->ptr2) 1156 if (res == tmpbuf->ptr)
1443 *res++ = '/'; 1157 *res++ = '/';
1444 1158
1445 req->result = res - (char *)req->ptr2; 1159 return res - (char *)tmpbuf->ptr;
1446 1160#endif
1447done:
1448 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1449} 1161}
1450 1162
1451static signed char 1163static signed char
1452eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1453{ 1165{
1461 1173
1462#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1174#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1463#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1464 1176
1465static void 1177static void
1466eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1178eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1467{ 1179{
1468 unsigned char bits [9 + sizeof (ino_t) * 8]; 1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1469 unsigned char *bit = bits; 1181 unsigned char *bit = bits;
1470 1182
1471 assert (CHAR_BIT == 8); 1183 assert (CHAR_BIT == 8);
1472 assert (sizeof (eio_dirent) * 8 < 256); 1184 assert (sizeof (eio_dirent) * 8 < 256);
1473 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1475 1187
1476 if (size <= EIO_SORT_FAST) 1188 if (size <= EIO_SORT_FAST)
1477 return; 1189 return;
1478 1190
1479 /* first prepare an array of bits to test in our radix sort */ 1191 /* first prepare an array of bits to test in our radix sort */
1480 /* try to take endianness into account, as well as differences in ino_t sizes */ 1192 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1481 /* inode_bits must contain all inodes ORed together */ 1193 /* inode_bits must contain all inodes ORed together */
1482 /* which is used to skip bits that are 0 everywhere, which is very common */ 1194 /* which is used to skip bits that are 0 everywhere, which is very common */
1483 { 1195 {
1484 ino_t endianness; 1196 eio_ino_t endianness;
1485 int i, j; 1197 int i, j;
1486 1198
1487 /* we store the byte offset of byte n into byte n of "endianness" */ 1199 /* we store the byte offset of byte n into byte n of "endianness" */
1488 for (i = 0; i < sizeof (ino_t); ++i) 1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1489 ((unsigned char *)&endianness)[i] = i; 1201 ((unsigned char *)&endianness)[i] = i;
1490 1202
1491 *bit++ = 0; 1203 *bit++ = 0;
1492 1204
1493 for (i = 0; i < sizeof (ino_t); ++i) 1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1494 { 1206 {
1495 /* shifting off the byte offsets out of "endianness" */ 1207 /* shifting off the byte offsets out of "endianness" */
1496 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1497 endianness >>= 8; 1209 endianness >>= 8;
1498 1210
1499 for (j = 0; j < 8; ++j) 1211 for (j = 0; j < 8; ++j)
1500 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1501 *bit++ = offs + j; 1213 *bit++ = offs + j;
1502 } 1214 }
1503 1215
1504 for (j = 0; j < 8; ++j) 1216 for (j = 0; j < 8; ++j)
1505 if (score_bits & (1 << j)) 1217 if (score_bits & (1 << j))
1506 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1507 } 1219 }
1508 1220
1509 /* now actually do the sorting (a variant of MSD radix sort) */ 1221 /* now actually do the sorting (a variant of MSD radix sort) */
1510 { 1222 {
1511 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1512 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1513 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1514 int stk_idx = 0; 1226 int stk_idx = 0;
1515 1227
1516 base_stk [stk_idx] = dents; 1228 base_stk [stk_idx] = dents;
1517 end_stk [stk_idx] = dents + size; 1229 end_stk [stk_idx] = dents + size;
1518 bit_stk [stk_idx] = bit - 1; 1230 bit_stk [stk_idx] = bit - 1;
1597 } 1309 }
1598 } 1310 }
1599} 1311}
1600 1312
1601static void 1313static void
1602eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1314eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1603{ 1315{
1604 if (size <= 1) 1316 if (size <= 1)
1605 return; /* our insertion sort relies on size > 0 */ 1317 return; /* our insertion sort relies on size > 0 */
1606 1318
1607 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1319 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1615 1327
1616/* read a full directory */ 1328/* read a full directory */
1617static void 1329static void
1618eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1619{ 1331{
1620 DIR *dirp;
1621 EIO_STRUCT_DIRENT *entp;
1622 char *name, *names; 1332 char *name, *names;
1623 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1624 int namesoffs = 0; 1334 int namesoffs = 0;
1625 int flags = req->int1; 1335 int flags = req->int1;
1626 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1627 int dentalloc = 128; 1337 int dentalloc = 128;
1628 int dentoffs = 0; 1338 int dentoffs = 0;
1629 ino_t inode_bits = 0; 1339 eio_ino_t inode_bits = 0;
1340#ifdef _WIN32
1341 HANDLE dirp;
1342 WIN32_FIND_DATA entp;
1343#else
1344 DIR *dirp;
1345 EIO_STRUCT_DIRENT *entp;
1346#endif
1630 1347
1631 req->result = -1; 1348 req->result = -1;
1632 1349
1633 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1634 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1635 1352
1636 X_LOCK (wrklock); 1353#ifdef _WIN32
1637 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1354 {
1355 int len = strlen ((const char *)req->ptr1);
1356 char *path = malloc (MAX_PATH);
1357 const char *fmt;
1358 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1359
1360 if (!len)
1361 fmt = "./*";
1362 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1363 fmt = "%s*";
1364 else
1365 fmt = "%s/*";
1366
1367 _snprintf (path, MAX_PATH, fmt, reqpath);
1368 dirp = FindFirstFile (path, &entp);
1369 free (path);
1370
1371 if (dirp == INVALID_HANDLE_VALUE)
1372 {
1373 /* should steal _dosmaperr */
1374 switch (GetLastError ())
1375 {
1376 case ERROR_FILE_NOT_FOUND:
1377 req->result = 0;
1378 break;
1379
1380 case ERROR_INVALID_NAME:
1381 case ERROR_PATH_NOT_FOUND:
1382 case ERROR_NO_MORE_FILES:
1383 errno = ENOENT;
1384 break;
1385
1386 case ERROR_NOT_ENOUGH_MEMORY:
1387 errno = ENOMEM;
1388 break;
1389
1390 default:
1391 errno = EINVAL;
1392 break;
1393 }
1394
1395 return;
1396 }
1397 }
1398#else
1399 #if HAVE_AT
1400 if (req->wd)
1401 {
1402 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1403
1404 if (fd < 0)
1405 return;
1406
1407 dirp = fdopendir (fd);
1408
1409 if (!dirp)
1410 silent_close (fd);
1411 }
1412 else
1638 self->dirp = dirp = opendir (req->ptr1); 1413 dirp = opendir (req->ptr1);
1414 #else
1415 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1416 #endif
1417
1418 if (!dirp)
1419 return;
1420#endif
1639 1421
1640 if (req->flags & EIO_FLAG_PTR1_FREE) 1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1641 free (req->ptr1); 1423 free (req->ptr1);
1642 1424
1643 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1644 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1645 req->ptr2 = names = malloc (namesalloc); 1427 req->ptr2 = names = malloc (namesalloc);
1646 X_UNLOCK (wrklock);
1647 1428
1648 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1649 for (;;) 1432 for (;;)
1650 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1651 errno = 0; 1439 errno = 0;
1652 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1653 1443
1654 if (!entp) 1444 if (done)
1655 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1656 if (errno) 1451 if (errno)
1657 break; 1452 break;
1453#endif
1658 1454
1659 /* sort etc. */ 1455 /* sort etc. */
1660 req->int1 = flags; 1456 req->int1 = flags;
1661 req->result = dentoffs; 1457 req->result = dentoffs;
1662 1458
1663 if (flags & EIO_READDIR_STAT_ORDER) 1459 if (flags & EIO_READDIR_STAT_ORDER)
1664 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1665 else if (flags & EIO_READDIR_DIRS_FIRST) 1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1666 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1667 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1463 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1668 else 1464 else
1465 {
1466 /* in this case, all is known, and we just put dirs first and sort them */
1467 eio_dirent *oth = dents + dentoffs;
1468 eio_dirent *dir = dents;
1469
1470 /* now partition dirs to the front, and non-dirs to the back */
1471 /* by walking from both sides and swapping if necessary */
1472 while (oth > dir)
1473 {
1474 if (dir->type == EIO_DT_DIR)
1475 ++dir;
1476 else if ((--oth)->type == EIO_DT_DIR)
1477 {
1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1479
1480 ++dir;
1481 }
1482 }
1483
1484 /* now sort the dirs only (dirs all have the same score) */
1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1486 }
1487
1488 break;
1489 }
1490
1491 /* now add the entry to our list(s) */
1492 name = D_NAME (entp);
1493
1494 /* skip . and .. entries */
1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1496 {
1497 int len = D_NAMLEN (entp) + 1;
1498
1499 while (ecb_expect_false (namesoffs + len > namesalloc))
1500 {
1501 namesalloc *= 2;
1502 req->ptr2 = names = realloc (names, namesalloc);
1503
1504 if (!names)
1505 break;
1506 }
1507
1508 memcpy (names + namesoffs, name, len);
1509
1510 if (dents)
1511 {
1512 struct eio_dirent *ent;
1513
1514 if (ecb_expect_false (dentoffs == dentalloc))
1669 { 1515 {
1670 /* in this case, all is known, and we just put dirs first and sort them */ 1516 dentalloc *= 2;
1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1518
1519 if (!dents)
1520 break;
1521 }
1522
1671 eio_dirent *oth = dents + dentoffs; 1523 ent = dents + dentoffs;
1672 eio_dirent *dir = dents;
1673 1524
1674 /* now partition dirs to the front, and non-dirs to the back */ 1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1675 /* by walking from both sides and swapping if necessary */ 1526 ent->namelen = len - 1;
1676 while (oth > dir) 1527 ent->inode = D_INO (entp);
1528
1529 inode_bits |= ent->inode;
1530
1531 switch (D_TYPE (entp))
1532 {
1533 default:
1534 ent->type = EIO_DT_UNKNOWN;
1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1536 break;
1537
1538 #ifdef DT_FIFO
1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1540 #endif
1541 #ifdef DT_CHR
1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1543 #endif
1544 #ifdef DT_MPC
1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1546 #endif
1547 #ifdef DT_DIR
1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1549 #endif
1550 #ifdef DT_NAM
1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1552 #endif
1553 #ifdef DT_BLK
1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1555 #endif
1556 #ifdef DT_MPB
1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1558 #endif
1559 #ifdef DT_REG
1560 case DT_REG: ent->type = EIO_DT_REG; break;
1561 #endif
1562 #ifdef DT_NWK
1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1564 #endif
1565 #ifdef DT_CMP
1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1567 #endif
1568 #ifdef DT_LNK
1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1570 #endif
1571 #ifdef DT_SOCK
1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1573 #endif
1574 #ifdef DT_DOOR
1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1576 #endif
1577 #ifdef DT_WHT
1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1579 #endif
1580 }
1581
1582 ent->score = 7;
1583
1584 if (flags & EIO_READDIR_DIRS_FIRST)
1585 {
1586 if (ent->type == EIO_DT_UNKNOWN)
1677 { 1587 {
1678 if (dir->type == EIO_DT_DIR) 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1679 ++dir; 1589 ent->score = 1;
1680 else if ((--oth)->type == EIO_DT_DIR) 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1681 { 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1682 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1683
1684 ++dir;
1685 }
1686 } 1592 }
1687 1593 else if (ent->type == EIO_DT_DIR)
1688 /* now sort the dirs only (dirs all have the same score) */ 1594 ent->score = 0;
1689 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1690 } 1595 }
1691
1692 break;
1693 }
1694
1695 /* now add the entry to our list(s) */
1696 name = entp->d_name;
1697
1698 /* skip . and .. entries */
1699 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1700 {
1701 int len = D_NAMLEN (entp) + 1;
1702
1703 while (ecb_expect_false (namesoffs + len > namesalloc))
1704 {
1705 namesalloc *= 2;
1706 X_LOCK (wrklock);
1707 req->ptr2 = names = realloc (names, namesalloc);
1708 X_UNLOCK (wrklock);
1709
1710 if (!names)
1711 break;
1712 } 1596 }
1713 1597
1714 memcpy (names + namesoffs, name, len);
1715
1716 if (dents)
1717 {
1718 struct eio_dirent *ent;
1719
1720 if (ecb_expect_false (dentoffs == dentalloc))
1721 {
1722 dentalloc *= 2;
1723 X_LOCK (wrklock);
1724 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1725 X_UNLOCK (wrklock);
1726
1727 if (!dents)
1728 break;
1729 }
1730
1731 ent = dents + dentoffs;
1732
1733 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1734 ent->namelen = len - 1;
1735 ent->inode = D_INO (entp);
1736
1737 inode_bits |= ent->inode;
1738
1739 switch (D_TYPE (entp))
1740 {
1741 default:
1742 ent->type = EIO_DT_UNKNOWN;
1743 flags |= EIO_READDIR_FOUND_UNKNOWN;
1744 break;
1745
1746 #ifdef DT_FIFO
1747 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1748 #endif
1749 #ifdef DT_CHR
1750 case DT_CHR: ent->type = EIO_DT_CHR; break;
1751 #endif
1752 #ifdef DT_MPC
1753 case DT_MPC: ent->type = EIO_DT_MPC; break;
1754 #endif
1755 #ifdef DT_DIR
1756 case DT_DIR: ent->type = EIO_DT_DIR; break;
1757 #endif
1758 #ifdef DT_NAM
1759 case DT_NAM: ent->type = EIO_DT_NAM; break;
1760 #endif
1761 #ifdef DT_BLK
1762 case DT_BLK: ent->type = EIO_DT_BLK; break;
1763 #endif
1764 #ifdef DT_MPB
1765 case DT_MPB: ent->type = EIO_DT_MPB; break;
1766 #endif
1767 #ifdef DT_REG
1768 case DT_REG: ent->type = EIO_DT_REG; break;
1769 #endif
1770 #ifdef DT_NWK
1771 case DT_NWK: ent->type = EIO_DT_NWK; break;
1772 #endif
1773 #ifdef DT_CMP
1774 case DT_CMP: ent->type = EIO_DT_CMP; break;
1775 #endif
1776 #ifdef DT_LNK
1777 case DT_LNK: ent->type = EIO_DT_LNK; break;
1778 #endif
1779 #ifdef DT_SOCK
1780 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1781 #endif
1782 #ifdef DT_DOOR
1783 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1784 #endif
1785 #ifdef DT_WHT
1786 case DT_WHT: ent->type = EIO_DT_WHT; break;
1787 #endif
1788 }
1789
1790 ent->score = 7;
1791
1792 if (flags & EIO_READDIR_DIRS_FIRST)
1793 {
1794 if (ent->type == EIO_DT_UNKNOWN)
1795 {
1796 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1797 ent->score = 1;
1798 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1799 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1800 }
1801 else if (ent->type == EIO_DT_DIR)
1802 ent->score = 0;
1803 }
1804 }
1805
1806 namesoffs += len; 1598 namesoffs += len;
1807 ++dentoffs; 1599 ++dentoffs;
1808 } 1600 }
1809 1601
1810 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1811 { 1603 {
1812 errno = ECANCELED; 1604 errno = ECANCELED;
1813 break; 1605 break;
1814 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1815 } 1613 }
1614#endif
1615 }
1816} 1616}
1617
1618/*****************************************************************************/
1619/* working directory stuff */
1620/* various deficiencies in the posix 2008 api force us to */
1621/* keep the absolute path in string form at all times */
1622/* fuck yeah. */
1623
1624#if !HAVE_AT
1625
1626/* a bit like realpath, but usually faster because it doesn'T have to return */
1627/* an absolute or canonical path */
1628static const char *
1629wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 if (!wd || *path == '/')
1632 return path;
1633
1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
1637 {
1638 int l1 = wd->len;
1639 int l2 = strlen (path);
1640
1641 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1642
1643 memcpy (res, wd->str, l1);
1644 res [l1] = '/';
1645 memcpy (res + l1 + 1, path, l2 + 1);
1646
1647 return res;
1648 }
1649}
1650
1651#endif
1652
1653static eio_wd
1654eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1655{
1656 int fd;
1657 eio_wd res;
1658 int len = eio__realpath (tmpbuf, wd, path);
1659
1660 if (len < 0)
1661 return EIO_INVALID_WD;
1662
1663#if HAVE_AT
1664 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1665
1666 if (fd < 0)
1667 return EIO_INVALID_WD;
1668#endif
1669
1670 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1671
1672#if HAVE_AT
1673 res->fd = fd;
1674#endif
1675
1676 res->len = len;
1677 memcpy (res->str, tmpbuf->ptr, len);
1678 res->str [len] = 0;
1679
1680 return res;
1681}
1682
1683eio_wd
1684eio_wd_open_sync (eio_wd wd, const char *path)
1685{
1686 struct etp_tmpbuf tmpbuf = { };
1687 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1688 free (tmpbuf.ptr);
1689
1690 return wd;
1691}
1692
1693void
1694eio_wd_close_sync (eio_wd wd)
1695{
1696 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1697 {
1698 #if HAVE_AT
1699 eio__close (wd->fd);
1700 #endif
1701 free (wd);
1702 }
1703}
1704
1705#if HAVE_AT
1706
1707static int
1708eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1709{
1710#if HAVE_RENAMEAT2
1711 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1712#else
1713 if (flags)
1714 return EIO_ENOSYS ();
1715
1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1717#endif
1718}
1719
1720/* they forgot these */
1721
1722static int
1723eio__truncateat (int dirfd, const char *path, off_t length)
1724{
1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1726 int res;
1727
1728 if (fd < 0)
1729 return fd;
1730
1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1734}
1735
1736static int
1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1738{
1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1740 int res;
1741
1742 if (fd < 0)
1743 return fd;
1744
1745 res = fstatvfs (fd, buf);
1746 silent_close (fd);
1747 return res;
1748}
1749
1750#endif
1817 1751
1818/*****************************************************************************/ 1752/*****************************************************************************/
1819 1753
1820#define ALLOC(len) \ 1754#define ALLOC(len) \
1821 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1822 { \ 1756 { \
1823 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1824 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1825 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1826 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1827 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1828 { \ 1762 { \
1829 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1830 req->result = -1; \ 1764 req->result = -1; \
1831 break; \ 1765 goto alloc_fail; \
1832 } \ 1766 } \
1833 } 1767 }
1834 1768
1835X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1836{
1837 ETP_REQ *req;
1838 struct timespec ts;
1839 etp_worker *self = (etp_worker *)thr_arg;
1840 int timeout;
1841 1770
1842 /* try to distribute timeouts somewhat evenly */ 1771static void
1843 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1844 1775
1845 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1846 { 1780 {
1847 ts.tv_sec = 0; 1781 off_t size = lseek (fd, 0, SEEK_END);
1848 1782
1849 X_LOCK (reqlock);
1850
1851 for (;;)
1852 {
1853 self->req = req = reqq_shift (&req_queue);
1854
1855 if (req)
1856 break;
1857
1858 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1859 {
1860 X_UNLOCK (reqlock);
1861 X_LOCK (wrklock);
1862 --started;
1863 X_UNLOCK (wrklock);
1864 goto quit;
1865 }
1866
1867 ++idle;
1868
1869 if (idle <= max_idle)
1870 /* we are allowed to idle, so do so without any timeout */
1871 X_COND_WAIT (reqwait, reqlock);
1872 else
1873 {
1874 /* initialise timeout once */
1875 if (!ts.tv_sec)
1876 ts.tv_sec = time (0) + idle_timeout;
1877
1878 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1879 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1880 }
1881
1882 --idle;
1883 }
1884
1885 --nready;
1886
1887 X_UNLOCK (reqlock);
1888
1889 if (req->type < 0) 1783 if (req->offs < 0)
1890 goto quit; 1784 req->offs += size;
1891 1785
1892 ETP_EXECUTE (self, req); 1786 if (!req->size)
1893 1787 req->size = size - req->offs;
1894 X_LOCK (reslock);
1895
1896 ++npending;
1897
1898 if (!reqq_push (&res_queue, req) && want_poll_cb)
1899 want_poll_cb ();
1900
1901 self->req = 0;
1902 etp_worker_clear (self);
1903
1904 X_UNLOCK (reslock);
1905 } 1788 }
1906 1789
1907quit: 1790 ALLOC (req->size);
1908 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1909 etp_worker_free (self);
1910 X_UNLOCK (wrklock);
1911 1792
1912 return 0; 1793 silent_close (fd);
1913}
1914 1794
1915/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1916 1798
1917int ecb_cold 1799int ecb_cold
1918eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1919{ 1801{
1920 return etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1803 eio_done_poll_cb = done_poll;
1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1921} 1806}
1922 1807
1923ecb_inline void 1808ecb_inline void
1924eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1925{ 1810{
1926 free (req); 1811 free (req);
1927} 1812}
1928 1813
1929#define REQ(rtype) \ 1814#define REQ(rtype) \
1930 eio_req *req; \ 1815 eio_req *req; \
1931 \ 1816 \
1932 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1933 if (!req) \ 1818 if (!req) \
1934 return 0; \ 1819 return 0; \
1948 { \ 1833 { \
1949 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1950 return 0; \ 1835 return 0; \
1951 } 1836 }
1952 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1953static void 1840static void
1954eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1955{ 1842{
1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1956 if (ecb_expect_false (EIO_CANCELLED (req))) 1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1957 { 1850 {
1958 req->result = -1; 1851 req->result = -1;
1959 req->errorno = ECANCELED; 1852 req->errorno = ECANCELED;
1960 return; 1853 return;
1961 } 1854 }
1962 1855
1856 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1857 {
1858 req->result = -1;
1859 req->errorno = ENOENT;
1860 return;
1861 }
1862
1863 if (req->type >= EIO_OPEN)
1864 {
1865 #if HAVE_AT
1866 dirfd = WD2FD (req->wd);
1867 #else
1868 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1869 #endif
1870 }
1871
1963 switch (req->type) 1872 switch (req->type)
1964 { 1873 {
1874 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1875 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1876 break;
1877 case EIO_WD_CLOSE: req->result = 0;
1878 eio_wd_close_sync (req->wd); break;
1879
1880 case EIO_SEEK: eio__lseek (req); break;
1965 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1966 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1967 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1968 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1969 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1970 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1971 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1972 1888
1889 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1890 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1891
1973 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1892 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1974 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1893 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1894
1895#if HAVE_AT
1975 1896
1976 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1977 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1978 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1900 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1901 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1902 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1903 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1904 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1905 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1906
1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1908 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1909 req->result = req->wd && SINGLEDOT (req->ptr1)
1910 ? rmdir (req->wd->str)
1911 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1912 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1913 case EIO_RENAME: req->result = eio__renameat2 (
1914 dirfd,
1915 /* complications arise because "." cannot be renamed, so we might have to expand */
1916 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1917 WD2FD ((eio_wd)req->int3),
1918 req->ptr2,
1919 req->int2
1920 );
1921 break;
1922 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1929 if (req->result == EIO_PATH_MAX)
1930 {
1931 req->result = -1;
1932 errno = ENAMETOOLONG;
1933 }
1934 break;
1935 case EIO_UTIME:
1936 case EIO_FUTIME:
1937 {
1938 struct timespec ts[2];
1939 struct timespec *times;
1940
1941 if (req->nv1 != -1. || req->nv2 != -1.)
1942 {
1943 ts[0].tv_sec = req->nv1;
1944 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1945 ts[1].tv_sec = req->nv2;
1946 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1947
1948 times = ts;
1949 }
1950 else
1951 times = 0;
1952
1953 req->result = req->type == EIO_FUTIME
1954 ? futimens (req->int1, times)
1955 : utimensat (dirfd, req->ptr1, times, 0);
1956 }
1957 break;
1958
1959#else
1960
1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1979 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1964 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1965 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1966 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1967 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1968 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1969 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1970
1971 case EIO_UNLINK: req->result = unlink (path ); break;
1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1974 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1975 case EIO_LINK: req->result = link (path , req->ptr2); break;
1976 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1977 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1978 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1979 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1980 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1981 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1982 if (req->result == EIO_PATH_MAX)
1983 {
1984 req->result = -1;
1985 errno = ENAMETOOLONG;
1986 }
1987 break;
1988
1989 case EIO_UTIME:
1990 case EIO_FUTIME:
1991 {
1992 struct timeval tv[2];
1993 struct timeval *times;
1994
1995 if (req->nv1 != -1. || req->nv2 != -1.)
1996 {
1997 tv[0].tv_sec = req->nv1;
1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1999 tv[1].tv_sec = req->nv2;
2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2001
2002 times = tv;
2003 }
2004 else
2005 times = 0;
2006
2007 req->result = req->type == EIO_FUTIME
2008 ? futimes (req->int1, times)
2009 : utimes (req->ptr1, times);
2010 }
2011 break;
2012
2013#endif
2014
2015 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2016 {
2017 ALLOC (req->result);
2018 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2019 }
2020 break;
2021
1980 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1981 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1982 2024
1983 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1984 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1985 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1986 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1987 2027
1988 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1989 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2028 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1990 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1991 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2029 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1992 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1993 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1994 2031
1995 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1996 case EIO_CLOSE: req->result = close (req->int1); break; 2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1997 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1998 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1999 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
2000 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2001 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2002 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2003 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2004 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2005
2006 case EIO_REALPATH: eio__realpath (req, self); break;
2007
2008 case EIO_READLINK: ALLOC (PATH_MAX);
2009 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2010
2011 case EIO_SYNC: req->result = 0; sync (); break; 2034 case EIO_SYNC: req->result = 0; sync (); break;
2012 case EIO_FSYNC: req->result = fsync (req->int1); break; 2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
2013 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2036 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2037 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2038 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2014 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2039 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2015 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2016 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2041 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2017 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2018 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2019 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2043 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2020 2044
2021 case EIO_READDIR: eio__scandir (req, self); break; 2045 case EIO_READDIR: eio__scandir (req, self); break;
2022 2046
2023 case EIO_BUSY: 2047 case EIO_BUSY:
2033 req->result = select (0, 0, 0, 0, &tv); 2057 req->result = select (0, 0, 0, 0, &tv);
2034 } 2058 }
2035#endif 2059#endif
2036 break; 2060 break;
2037 2061
2038 case EIO_UTIME: 2062#if 0
2039 case EIO_FUTIME:
2040 {
2041 struct timeval tv[2];
2042 struct timeval *times;
2043
2044 if (req->nv1 != -1. || req->nv2 != -1.)
2045 {
2046 tv[0].tv_sec = req->nv1;
2047 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2048 tv[1].tv_sec = req->nv2;
2049 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2050
2051 times = tv;
2052 }
2053 else
2054 times = 0;
2055
2056 req->result = req->type == EIO_FUTIME
2057 ? futimes (req->int1, times)
2058 : utimes (req->ptr1, times);
2059 }
2060 break;
2061
2062 case EIO_GROUP: 2063 case EIO_GROUP:
2063 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
2064 2066
2065 case EIO_NOP: 2067 case EIO_NOP:
2066 req->result = 0; 2068 req->result = 0;
2067 break; 2069 break;
2068 2070
2069 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
2070 req->feed (req); 2072 req->feed (req);
2071 break; 2073 break;
2072 2074
2073 default: 2075 default:
2074 errno = ENOSYS;
2075 req->result = -1; 2076 req->result = EIO_ENOSYS ();
2076 break; 2077 break;
2077 } 2078 }
2078 2079
2080alloc_fail:
2079 req->errorno = errno; 2081 req->errorno = errno;
2080} 2082}
2081 2083
2082#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
2083 2085
2086eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_OPEN); PATH; SEND;
2089}
2090
2091eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2094}
2095
2084eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
2085{ 2097{
2086 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
2087} 2099}
2088 2100
2104eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2116eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2105{ 2117{
2106 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2118 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2107} 2119}
2108 2120
2121eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2124}
2125
2126eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2129}
2130
2131eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2134}
2135
2109eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2136eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2110{ 2137{
2111 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2138 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2112} 2139}
2113 2140
2119eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2120{ 2147{
2121 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2122} 2149}
2123 2150
2124eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2127}
2128
2129eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2130{ 2152{
2131 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2132} 2154}
2133 2155
2134eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2135{
2136 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2137}
2138
2139eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2140{ 2157{
2141 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2142} 2159}
2143 2160
2144eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2161eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2145{ 2162{
2146 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2163 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2147} 2164}
2148 2165
2166eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2169}
2170
2149eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2171eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2150{ 2172{
2151 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2173 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2152} 2174}
2153 2175
2154eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2176eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2155{ 2177{
2156 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2178 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2179}
2180
2181eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2184}
2185
2186eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2157} 2189}
2158 2190
2159eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2191eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2160{ 2192{
2161 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2306eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2338eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2307{ 2339{
2308 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2309} 2341}
2310 2342
2343eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2344{
2345 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2346}
2347
2311eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) 2348eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2312{ 2349{
2313 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2314} 2351}
2315 2352
2349void 2386void
2350eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
2351{ 2388{
2352 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2389 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2353 2390
2354 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
2355 2392
2356 ++grp->size; 2393 ++grp->size;
2357 req->grp = grp; 2394 req->grp = grp;
2358 2395
2359 req->grp_prev = 0; 2396 req->grp_prev = 0;
2369/* misc garbage */ 2406/* misc garbage */
2370 2407
2371eio_ssize_t 2408eio_ssize_t
2372eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2409eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2373{ 2410{
2374 etp_worker wrk;
2375 eio_ssize_t ret;
2376
2377 wrk.dbuf = 0;
2378
2379 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
2380
2381 if (wrk.dbuf)
2382 free (wrk.dbuf);
2383
2384 return ret;
2385} 2412}
2386 2413

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines